LiberAI's Liber-0 performs new tasks without task-specific demonstrations

LiberAI announced that its Liber-0 model achieved a 36.0% overall success rate across 60 GPT-6-proposed tasks without being trained on task-specific robot demonstrations. The model bridges the gap between human video experience and robot execution through action grounding and embodiment alignment. Pretrained on over 2.2 million hours of human experience, the system aims to build robot capabilities using fewer task-specific demonstrations. Open article →